DOCSLIB.ORG

4- Markov Chains and MCMC

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
4- Markov Chains and MCMC Markov Chains and MCMC Markov chains I Let S = f1; 2;:::; Ng be a finite set consisting of N states. I A Markov chain Y0; Y1; Y2;::: is a sequence of random variables, with Yt 2 S for all points in time t 2 N, that satisfies the Markov property, namely, given the present state, the future and past states are independent: Pr(Yn+1 = xjY0 = y0;:::; Yn = yn) = Pr(Yn+1 = xjYn = yn) I A Markov chain is homogeneous if for all n ≥ 1: Pr(Yn+1 = jjYn = i) = Pr(Yn = jjYn−1 = i) i.e., transitional probabilities are time independent. I There is an N × N transition matrix P such that for all i; j; n VPN we have: Pij := Pr(Yn+1 = jjYn = i) ≥ 0 j=1 Pij = 1. n×n PN I Any matrix P 2 R with Pij ≥ 0 and j=1 Pij = 1, for 1 ≤ i; j ≤ N, is called a stochastic matrix. Example I Let S = f1; 2; 3; 4g and 0 1=2 1=2 0 0 1 B 1=2 1=2 0 0 C P = B C (1) @ 1=3 1=6 1=6 1=3 A 0 0 0 1 I Here is the graphical representation of P, i.e., only the edges with Pij > 0 are drawn: 1/2 1/2 1/2 11/2 2 1/3 1/6 3 1/3 4 1 1/6 Markov chains as Dynamical Systems I A homogeneous Markov chain defines a dynamical system: N PN I Let X = fx 2 R : xi ≥ 0; i=1 xi = 1g: the space of probability vectors over S. 1 PN 1 I d(x; y) := 2 k=1 jxk − yk j = 2 kx − yk1 is a metric on X. I P : X ! X defines a linear map by right multiplication. PN I P : x 7! xP, i.e., (xP)n = k=1 xk Pkn, is well-defined: PN PN PN PN PN I n=1(xP)n = n=1 k=1 xk Pkn = k=1 n=1 xk Pkn = PN PN PN k=1 xk n=1 Pkn = k=1 xk = 1 (n) n I The n-step transition matrix is simply P = P . I If x 2 X then the probability vector over S evolves as orbit of x : x; xP; xP2;:::; xPn;::: I Interested in the long term behaviour of orbits. Communicating Classes I A state j is accessible from i, denoted by i ! j, if there (n) exists n ≥ 0 with Pij > 0, i.e., j can be reached from i after a finite number of steps. I Two states i and j communicate, denoted by i $ j, if i ! j and j ! i. I Note that for any state i, we have i $ i. Why? I Communication is an equivalence relation, which induces communicating classes. I By analysing the transition matrix, we determine the communication classes. I Find the communicating classes of P in Equation (1). I A Markov chain is irreducible if it has a single communicating class, i.e., if any state can be reached from any state. Aperiodic Markov Chains I The period of a state i is defined as (n) d(i) := gcdfn ≥ 1 : Pii > 0g. I Here, gcd denotes the greatest common divisor. I Example: Consider a Markov Chain with transition matrix, 0 0 1 0 0 1 B 1=2 0 1=2 0 C P = B C (2) @ 0 1=2 0 1=2 A 0 0 1 0 All diagonal elements of P2n are positive and those of P2n+1 are all zero. Thus, every state has period 2. I If d(i) = 1 then i is said to be aperiodic. I If all states are aperiodic then the chain is called aperiodic. Irreducibility and Periodicity: basic properties I If a Markov chain is irreducible then all its states have the same period. I An irreducible Markov chain is aperiodic if there is a state i with Pii > 0. I An irreducible Markov chain is aperiodic iff there exists (n) n ≥ 1 such that 8i; j:Pij > 0. I If P is the transition matrix of an irreducible Markov chain and 0 < a < 1, then aI + (1 − a)P is the transition matrix of an irreducible Markov chain, where I is the N × N identity matrix. I Thus, by choosing a small a > 0, we can allow a small self transition to make the Markov chain aperiodic. I Moreover, we will see later that P and aI + (1 − a)P have the same stationary distribution. Stationary distribution I A probability vector π 2 X is a stationary distribution of a Markov chain if πP = π. I Any transition matrix has an eigenvalue equal to 1. Why? I Thus, find the stationary distributions by solving the eigenvalue problem πP = π or PT πT = πT . I Fundamental Theorem of Markov chains. An irreducible and aperiodic Markov chain has a unique stationary distribution π which satisfies: n I limn!1 xP = π for all x 2 X. n I limn!1 P exists and is the matrix with all rows equal to π. I This means that whatever our starting probability vector, the dynamics of the chain takes it to the unique stationary distribution or the steady-state of the system. I Thus, we have an attractor π with basin X. I Check: πP = π () π(aI + (1 − a)P) = π, for 0 < a < 1. Reversible Chains and Detailed Balanced Condition I A Markov chain is reversible if there is π 2 X that satisfies the detailed balanced condition: πi Pij = πj Pji ; for 1 ≤ i; j ≤ N I This means that Pr(Yn = j; Yn−1 = i) = Pr(Yn = i; Yn−1 = j) when Pr(Yn−1 = i) = πi for all i, i.e., the chain is time reversible. I Exercise: If π satisfies the detailed balanced condition, then it is a stationary distribution. I This shows that satisfying the detailed balanced condition is sufficient for π to be the stationary distribution. I However, the detailed balanced condition is not a necessary condition for π to be a stationary distribution. Markov Chain Monte Carlo I Markov Chain Monte Carlo (MCMC) methods are based on the convergence of the orbit of any initial probability vector to the unique stationary distribution of an irreducible and aperiodic Markov chain. I If we need to sample from a distribution q say on a finite state space, we construct an irreducible and aperiodic Markov chain P with unique stationary distribution π = q. n n I Then, since limn!1 xP = π, i.e. limn!1 jxP − πj = 0 for any x 2 X, it follows that for large n, we have xPn ≈ π = q. 3 xP 4 x xP 2 xP n q xP xP I Metropolis-Hastings algorithms such as Gibbs sampling used in stochastic Hopfield networks and RBMs are examples of MCMC, as we will see. Stochastic Hopfield Networks I Replace the deterministic updating xi sgn(hi ), where PN hi = j=1 wij xj , with an asynchronous probabilistic rule: 1 1 Pr(xi = ±1) = = (3) 1 + exp(∓2hi =T ) 1 + exp(−2hi xi =T ) where T > 0 is the pseudo temperature. I As T ! 0, this is reduced to the deterministic rule. T=0 sgn 1 1 1/2 1 1 x −x/T 2 Logistic sigmoid map 1 + e T= 1 1/2 x 1/2 −x 1 + e 0 I Exercise. From (3), obtain the probability of flipping: 1 Pr(x ! −x ) = ; (4) i i 1 + exp(∆E=T ) where ∆E = E 0 − E is the change in energy. I There is now some probability that the energy increases. Stationary Distribution I A stochastic Hopfield network can be viewed as a Markov N process with 2 states: xi = ±1 for 1 ≤ i ≤ N. I The flipping probability (4) defines the transition matrix P. I Irreducible: The flipping rule is applied asynchronously at all nodes, one at a time, and therefore, there is a non-zero probability of going from any state to any state. I Aperiodic: There is a non-zero probability that at any time a state does not change. I Thus, the network has a unique stationary distribution. I Exercise: Show that the distribution exp(−E(x)=T ) π(x) := Pr(x) = ; (5) Z where Z is the normalisation factor, satisfies the detailed balanced condition, i.e., it is the stationary distribution. I Start with any configuration of the network; successively apply the flipping rule. After a large number of iterations, we have a sample from the stationary distribution. Computing average values I Suppose now we are interested to find not a sample from the stationary distribution π but the average value of a function f : X ! R on the configurations of a stochastic Hopfield network with respect to π. PN I For example, we may take f (x) = i=1 xi and want to PN compute Eπ( i=1 xi ), i.e., the average value of the sum of all binary node values with respect to the stationary distribution π of the stochastic network. I As we have seen, we can find a sample of π by starting with any distribution p0 and applying the transitional n probability P a large number of times, n say, to obtain p0P as a sample of π. I We thus reasonably expect that we can find a good PN estimate for Eπ( i=1 xi ) by computing the expected value N n N P n P of i=1 xi with respect to p0P , i.e., find Ep0P ( i=1 xi ).
Load more

Recommended publications
  • A Relativistic Extension of Hopfield Neural Networks Via the Mechanical Analogy
    A relativistic extension of Hopfield neural networks via the mechanical analogy Adriano Barra,a;b;c Matteo Beccariaa;b and Alberto Fachechia;b aDipartimento di Matematica e Fisica Ennio De Giorgi, Universit`adel Salento, Lecce, Italy bINFN, Istituto Nazionale di Fisica Nucleare, Sezione di Lecce, Italy cGNFM-INdAM, Gruppo Nazionale per la Fisica Matematica, Sezione di Lecce, Italy E-mail: [email protected], [email protected], [email protected] Abstract: We propose a modification of the cost function of the Hopfield model whose salient features shine in its Taylor expansion and result in more than pairwise interactions with alternate signs, suggesting a unified framework for handling both with deep learning and network pruning. In our analysis, we heavily rely on the Hamilton-Jacobi correspon- dence relating the statistical model with a mechanical system. In this picture, our model is nothing but the relativistic extension of the original Hopfield model (whose cost function is a quadratic form in the Mattis magnetization which mimics the non-relativistic Hamiltonian for a free particle). We focus on the low-storage regime and solve the model analytically by taking advantage of the mechanical analogy, thus obtaining a complete characterization of the free energy and the associated self-consistency equations in the thermodynamic limit. On the numerical side, we test the performances of our proposal with MC simulations, showing that the stability of spurious states (limiting the capabilities of the standard Heb- bian
    [Show full text]
  • Optimization by Mean Field Annealing
    91 OPTIMIZATION BY MEAN FIELD ANNEALING Griff Bilbro Reinhold Mann Thomas K. Miller ECE Dept. Eng. Physics and Math. Div. ECE Dept. NCSU Oak Ridge N atl. Lab. NCSU Raleigh, NC 27695 Oak Ridge, TN 37831 Raleigh, N C 27695 Wesley. E. Snyder David E. Van den Bout Mark White ECE Dept. ECE Dept. ECE Dept. NCSU NCSU NCSU Raleigh, NC 27695 Raleigh, NC 27695 Raleigh, NC 27695 ABSTRACT Nearly optimal solutions to many combinatorial problems can be found using stochastic simulated annealing. This paper extends the concept of simulated annealing from its original formulation as a Markov process to a new formulation based on mean field theory. Mean field annealing essentially replaces the discrete de­ grees of freedom in simulated annealing with their average values as computed by the mean field approximation. The net result is that equilibrium at a given temperature is achieved 1-2 orders of magnitude faster than with simulated annealing. A general frame­ work for the mean field annealing algorithm is derived, and its re­ lationship to Hopfield networks is shown. The behavior of MFA is examined both analytically and experimentally for a generic combi­ natorial optimization problem: graph bipartitioning. This analysis indicates the presence of critical temperatures which could be im­ portant in improving the performance of neural networks. STOCHASTIC VERSUS MEAN FIELD In combinatorial optimization problems, an objective function or Hamiltonian, H(s), is presented which depends on a vector of interacting 3pim, S = {81," .,8N}, in some complex nonlinear way. Stochastic simulated annealing (SSA) (S. Kirk­ patrick, C. Gelatt, and M. Vecchi (1983)) finds a global minimum of H by com­ bining gradient descent with a random process.
    [Show full text]
  • 7Network Models
    7 Network Models 7.1 Introduction Extensive synaptic connectivity is a hallmark of neural circuitry. For ex- ample, a typical neuron in the mammalian neocortex receives thousands of synaptic inputs. Network models allow us to explore the computational potential of such connectivity, using both analysis and simulations. As illustrations, we study in this chapter how networks can perform the fol- lowing tasks: coordinate transformations needed in visually guided reach- ing, selective amplification leading to models of simple and complex cells in primary visual cortex, integration as a model of short-term memory, noise reduction, input selection, gain modulation, and associative mem- ory. Networks that undergo oscillations are also analyzed, with applica- tion to the olfactory bulb. Finally, we discuss network models based on stochastic rather than deterministic dynamics, using the Boltzmann ma- chine as an example. Neocortical circuits are a major focus of our discussion. In the neocor- tex, which forms the convoluted outer surface of the (for example) human brain, neurons lie in six vertical layers highly coupled within cylindrical columns. Such columns have been suggested as basic functional units, cortical columns and stereotypical patterns of connections both within a column and be- tween columns are repeated across cortex. There are three main classes of interconnections within cortex, and in other areas of the brain as well. Feedforward connections bring input to a given region from another re- feedforward, gion located at an earlier stage along a particular processing pathway. Re- recurrent, current synapses interconnect neurons within a particular region that are and top-down considered to be at the same stage along the processing pathway.
    [Show full text]
  • Chapter 14 Stochastic Networks
    14 Stochastic Networks 14.1 Variations of the Hopfield model In the previous chapter we showed that Hopfield networks can be used to provide solutions to combinatorial problems that can be expressed as the minimization of an energy function, although without guaranteeing global optimality. Once the weights of the edges have been defined, the network shows spontaneous computational properties. Harnessing this spontaneous dynamics for useful computations requires some way of avoiding falling into local minima of the energy function in such a way that the global minimum is reached. In the case of the eight queens problem, the number of local minima is much higher than the number of global minima and very often the Hopfield network does not stabilize at a correct solution. The issue to be investigated is therefore whether a certain variation of the Hopfield model could achieve better results in combinatorial optimization problems. One possible strategy to improve the global convergence properties of a network consists in increasing the number of paths in search space, in such a way that not only binary but also real-valued states become possible. Contin- uous updates of the network state become possible and the search for minima of the energy function is no longer limited to the corners of a hypercube in state space. All interior points are now legal network states. A continuous acti- vation function also makes it possible to describe more precisely the electrical circuits used to implement Hopfield networks. A second strategy to avoid local minima of the energy function consists in introducing noise into the network dynamics.
    [Show full text]
  • Statistical Computations Underlying the Dynamics of Memory Updating
    Statistical Computations Underlying the Dynamics of Memory Updating Samuel J. Gershman1*, Angela Radulescu2, Kenneth A. Norman2, Yael Niv2 1 Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachussetts, United States of America, 2 Department of Psychology and Princeton Neuroscience Institute, Princeton University, Princeton, New Jersey, United States of America Abstract Psychophysical and neurophysiological studies have suggested that memory is not simply a carbon copy of our experience: Memories are modified or new memories are formed depending on the dynamic structure of our experience, and specifically, on how gradually or abruptly the world changes. We present a statistical theory of memory formation in a dynamic environment, based on a nonparametric generalization of the switching Kalman filter. We show that this theory can qualitatively account for several psychophysical and neural phenomena, and present results of a new visual memory experiment aimed at testing the theory directly. Our experimental findings suggest that humans can use temporal discontinuities in the structure of the environment to determine when to form new memory traces. The statistical perspective we offer provides a coherent account of the conditions under which new experience is integrated into an old memory versus forming a new memory, and shows that memory formation depends on inferences about the underlying structure of our experience. Citation: Gershman SJ, Radulescu A, Norman KA, Niv Y (2014) Statistical Computations Underlying the Dynamics of Memory Updating. PLoS Comput Biol 10(11): e1003939. doi:10.1371/journal.pcbi.1003939 Editor: Olaf Sporns, Indiana University, United States of America Received August 19, 2013; Accepted September 26, 2014; Published November 6, 2014 Copyright: ß 2014 Gershman et al.
    [Show full text]
  • Forecast of Consumer Behaviour Based on Neural Networks Models Comparison
    ACTA UNIVERSITATIS AGRICULTURAE ET SILVICULTURAE MENDELIANAE BRUNENSIS Volume LX 57 Number 2, 2012 FORECAST OF CONSUMER BEHAVIOUR BASED ON NEURAL NETWORKS MODELS COMPARISON M. Štencl, O. Popelka, J. Šťastný Received: November 30, 2011 Abstract ŠTENCL, POPELKA, ŠŤASTNÝ: Forecast of consumer behaviour based on neural networks models comparison. Acta univ. agric. et silvic. Mendel. Brun., 2012, LX, No. 2, pp. 437–442 The aim of this article is comparison of accuracy level of forecasted values of several artifi cial neural network models. The comparison is performed on datasets of Czech household consumption values. Several statistical models o en resolve this task with more or fewer restrictions. In previous work where models’ input conditions were not so strict and model with missing data was used (the time series didn’t contain many values) we have obtained comparably good results with artifi cial neural networks. Two views – practical and theoretical, motivate the purpose of this study. Forecasting models for medium term prognosis of the main trends of Czech household consumption is part of the faculty research design grant MSM 6215648904/03/02 (Sub-task 5.3) which defi nes the practical purpose. Testing of nonlinear autoregressive artifi cial neural network model compared with feed- forward neural network and radial basis function neural network defi nes the theoretical purpose. The performance metrics of the models were evaluated using a combination of common error metrics, namely Correlation Coeffi cient and Mean Square Error, together with the number of epochs and/or main prediction error. artifi cial neural networks, forecasting methods, customer behaviour Prediction of tendencies in consumer behaviour of statistical learning, but they are built on diff erent of Czech (or any other) households is one of key fundamentals.
    [Show full text]
  • Artificial Neural Networks for Financial Time Series Prediction and Portfolio
    Master of Science Thesis in Industrial Engineering and Management Department of Management and Engineering, Linköping University June 2017 Artificial neural networks for financial time series prediction and portfolio optimization Samuel Björklund Tobias Uhlin Credits: 30HP Level: A Advisor: Jörgen Blomvall Department of Management and Engineering, Linköping University Examiner: Ou Tang Department of Management and Engineering, Linköping University ISRN: LIU-IEI-TEK-A--17/02920—SE Abstract Predicting the return of financial times series using traditional technical analysis and widely used economic models such as the capital asset pricing model, has proven to be difficult. Machine learning is a subfield of computer science, a concept that is frequently being used within different domains and recurrently delivers successful results. Artificial neural network is a product from the field of machine learning, a black box model that if properly designed processes data, learns its dynamic and subsequently provides an informative output. The objective of this thesis is to develop artificial neural networks that predict time varying expected return of financial time series with purpose of optimizing portfolio weights. This thesis begins with a review of prevailing characteristics of financial times series, and existing methods for estimating statistical properties. Background on artificial neural network along with empirical case studies of their suitability within the fi- nancial domain is put into a brief discussion vis-a-vis the efficient market hypothesis where potential anomalies is highlighted. Based on the theoretical review, an inter- disciplinary approach between machine learning and finance culminated into a model that estimates the expected return of time series on a quarterly basis.
    [Show full text]
  • Online Discrimination of Nonlinear Dynamics with Switching Differential Equations
    Online Discrimination of Nonlinear Dynamics with Switching Differential Equations Sebastian Bitzer, Izzet B. Yildiz and Stefan J. Kiebel Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig fbitzer,yildiz,[email protected] Abstract How to recognise whether an observed person walks or runs? We consider a dy- namic environment where observations (e.g. the posture of a person) are caused by different dynamic processes (walking or running) which are active one at a time and which may transition from one to another at any time. For this setup, switch- ing dynamic models have been suggested previously, mostly, for linear and non- linear dynamics in discrete time. Motivated by basic principles of computations in the brain (dynamic, internal models) we suggest a model for switching non- linear differential equations. The switching process in the model is implemented by a Hopfield network and we use parametric dynamic movement primitives to represent arbitrary rhythmic motions. The model generates observed dynamics by linearly interpolating the primitives weighted by the switching variables and it is constructed such that standard filtering algorithms can be applied. In two experi- ments with synthetic planar motion and a human motion capture data set we show that inference with the unscented Kalman filter can successfully discriminate sev- eral dynamic processes online. 1 Introduction Humans are extremely accurate when discriminating rapidly between different dynamic processes occurring in their environment. For example, for us it is a simple task to recognise whether an observed person is walking or running and we can use subtle cues in the structural and dynamic pattern of an observed movement to identify emotional state, gender and intent of a person.
    [Show full text]
  • On the Mapping Between Hopfield Networks and Restricted Boltzmann
    Published as a conference paper at ICLR 2021 ON THE MAPPING BETWEEN HOPFIELD NETWORKS AND RESTRICTED BOLTZMANN MACHINES Matthew Smart Anton Zilman Department of Physics Department of Physics University of Toronto and Institute for Biomedical Engingeering [email protected] University of Toronto [email protected] ABSTRACT Hopfield networks (HNs) and Restricted Boltzmann Machines (RBMs) are two important models at the interface of statistical physics, machine learning, and neu- roscience. Recently, there has been interest in the relationship between HNs and RBMs, due to their similarity under the statistical mechanics formalism. An exact mapping between HNs and RBMs has been previously noted for the special case of orthogonal (“uncorrelated”) encoded patterns. We present here an exact map- ping in the case of correlated pattern HNs, which are more broadly applicable to existing datasets. Specifically, we show that any HN with N binary variables and p < N arbitrary binary patterns can be transformed into an RBM with N binary visible variables and p gaussian hidden variables. We outline the conditions under which the reverse mapping exists, and conduct experiments on the MNIST dataset which suggest the mapping provides a useful initialization to the RBM weights. We discuss extensions, the potential importance of this correspondence for the training of RBMs, and for understanding the performance of deep architectures which utilize RBMs. 1 INTRODUCTION Hopfield networks (HNs) (Hopfield, 1982; Amit, 1989) are a classical neural network architecture that can store prescribed patterns as fixed-point attractors of a dynamical system. In their standard formulation with binary valued units, HNs can be regarded as spin glasses with pairwise interactions Jij that are fully determined by the patterns to be encoded.
    [Show full text]
  • Generalization in a Hopfield Network J.F
    Generalization in a Hopfield network J.F. Fontanari To cite this version: J.F. Fontanari. Generalization in a Hopfield network. Journal de Physique, 1990, 51 (21), pp.2421- 2430. 10.1051/jphys:0199000510210242100. jpa-00212540 HAL Id: jpa-00212540 https://hal.archives-ouvertes.fr/jpa-00212540 Submitted on 1 Jan 1990 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. J. Phys. France 51 (1990) 2421-2430 ler NOVEMBRE 1990, 2421 Classification Physics Abstracts 87.10 - 64.60C Generalization in a Hopfield network J. F. Fontanari Instituto de Fisica e Quimica de São Carlos, Universidade de São Paulo, Caixa Postal 369, 13560 São Carlos SP, Brasil (Received 23 April 1990, accepted in final form 12 July 1990) Abstract. 2014 The performance of a Hopfield network in learning an extensive number of concepts having access only to a finite supply of typical data which exemplify the concepts is studied. The minimal number of examples which must be taught to the network in order it starts to create representations for the concepts is calculated analitically. It is shown that the mixture states play a crucial role in the creation of these representations.
    [Show full text]
  • A Novel Neural Network Training Framework with Data Assimilation
    A Novel Neural Network Training Framework with Data Assimilation Chong Chen1, *, Qinghui Xing1, Xin Ding1, Yaru Xue1, Tianfu Zhong1 1 College of Information Science and Engineering, China University of Petroleum – Beijing * Email: [email protected] Abstract In recent years, the prosperity of deep learning has revolutionized the Artificial Neural Networks. However, the dependence of gradients and the offline training mechanism in the learning algorithms prevents the ANN for further improvement. In this study, a gradient-free training framework based on data assimilation is proposed to avoid the calculation of gradients. In data assimilation algorithms, the error covariance between the forecasts and observations is used to optimize the parameters. Feedforward Neural Networks (FNNs) are trained by gradient decent, data assimilation algorithms (Ensemble Kalman Filter (EnKF) and Ensemble Smoother with Multiple Data Assimilation (ESMDA)), respectively. ESMDA trains FNN with pre-defined iterations by updating the parameters using all the available observations which can be regard as offline learning. EnKF optimize FNN when new observation available by updating parameters which can be regard as online learning. Two synthetic cases with the regression of a Sine Function and a Mexican Hat function are assumed to validate the effectiveness of the proposed framework. The Root Mean Square Error (RMSE) and coefficient of determination (R2) are used as criteria to assess the performance of different methods. The results show that the proposed training framework performed better than the gradient decent method. The proposed framework provides alternatives for online/offline training the existing ANNs (e.g., Convolutional Neural Networks, Recurrent Neural Networks) without the dependence of gradients.
    [Show full text]
  • Multi-Agent Market Modeling Based on Neural Networks
    MULTI-AGENT MARKET MODELING BASED ON NEURAL NETWORKS by RALPH GROTHMANN University of Bremen, Germany Siemens AG, Corporate Technology, Munich, Germany Thesis presented for the Degree of Doctor of Economics (Dr. rer. pol.) Faculty of Economics, University of Bremen, Germany Contents Acronyms xi List of Figures xiii List of Tables xvii 1 Introduction 1 2 Neural Networks: Introduction & Historical Notes 15 1. From Biological to Artificial Neurons 16 1.1 Biological Neurons 16 1.2 Artificial Neurons 18 1.3 Common Activation Functions 20 1.4 Neurons as Elementary Models of Decision Making 21 2. 3-Layer Feedforward Networks 23 2.1 Architecture of the Standard Model 24 2.2 Universal Functional Approximation Abilities 26 2.3 Representation of Temporal Structures 26 2.4 Input Preprocessing 27 2.5 The Dilemma of Overfitting 28 2.6 Early Stopping 29 2.7 Late Stopping 30 3. Historical Notes on Neural Networks 32 3.1 McCulloch and Pitts Model of a Neuron (1943) 33 3.2 Hebbian Learning Rule (1949) 33 3.3 Perceptron of Rosenblatt (1958) 34 3.4 Adaptive Linear Element (1960) 36 3.5 Minsky and Papert's Critics of Perceptrons (1969) 36 3.6 Self-Organizing Maps (1982) 37 3.7 Hopfield Networks (1982) 38 3.8 Boltzmann Machine (1985) 39 3.9 Standard Error Backpropagation (1974 & 1986) 41 v vi MULTI-AGENT MARKET MODELING BY NEURAL NETWORKS 3.10 Radial Basis Function Networks (1988) 42 3.11 Elman's Recurrent Neural Networks (1990) 42 3.12 Dilemma of Overfitting & Purely Data Driven Modeling 45 4. Preliminary Conclusion 46 3 Modeling Dynamical Systems by Feedforward Neural Networks 49 1.
    [Show full text]